Optimizing Stability for Intel 13900k and 14900k CPU’s

[Panzerlang]

It might already have been posted but if not:
 

 

[SharpeXB]

I did actually just solve a bout of CDT fails with games by updating my ASUS mb BIOS. I’m not enough of an expert to know why this worked exactly or why it was necessary but it solved the problem. I notice the new default settings disable MCE. I did have this disabled prior but had to do that manually. 

Edited July 31, 2024 by SharpeXB

[LucShep]

Just by activating "Synch All Cores" in your P-Cores and placing a value (for all the cores) that is same or as close to the "out of the box" (Stock) all-core clock for the specific processor, you imediately fix the worst part of the problem.

What degrades these CPUs faster is when a single core, even for a background task, asks to boost to whatever outrageous ammount of "single/dual core boost".
Even idling in desktop you can see this destructive behaviour, doing its own unknowningly suicidal thing.
It happens very, very frequently during whatever time of use.

When it does that, it is given 1.50v+ depending on the individual CPU and motherboard+bios (some people even mention 1.60v at times).
That is one major reason why these chips are dying. Even though the chip has power limits, when only one core wants to boost, ALL of the power is offered, at once.
The most insane part is that this seems to be a feature just so Intel (and also AMD has it) can get a good CineBench score by reviewers.

Lock those cores and you'll be a LOT better.

Edited July 31, 2024 by LucShep

[Panzerlang]

4 hours ago, LucShep said:

Just by activating "Synch All Cores" and placing a value (for all the cores) that is same or as close to the "out of the box" (Stock) all-core clock for the specific processor, you imediately fix the worst part of the problem.

What degrades these CPUs faster is when a single core, even for a background task, asks to boost to whatever outrageous ammount of "single/dual core boost".
Even idling in desktop you can see this destructive behaviour, doing its own unknowningly suicidal thing.
It happens very, very frequently during whatever time of use.

When it does that, it is given 1.50v+ depending on the individual CPU and motherboard+bios (some people even mention 1.60v at times).
That is one major reason why these chips are dying. Even though the chip has power limits, when only one core wants to boost, ALL of the power is offered, at once.
The most insane part is that this seems to be a feature just so Intel (and also AMD has it) can get a good CineBench score by reviewers.

Lock those cores and you'll be a LOT better.

 

Doesn't locking the cores also force them to run at max speed permanently?

[The_Nephilim]

17 minutes ago, Panzerlang said:

Doesn't locking the cores also force them to run at max speed permanently?

well if you run speedstep it will declock when just idle or doing minimal work.. if you do not want it running max all the time..

[LucShep]

On 7/31/2024 at 10:20 PM, Panzerlang said:

Doesn't locking the cores also force them to run at max speed permanently?

If you haven't disabled the CPU power savings (Speedstep, Speedshift, and all the C-States), then you're just saying to the motherboard that "I want those P-cores to all be limited to that X ammount of Ghz, when at maximum".  It'll still downclock like always, power savings the same, everything the same, when you're not pushing the processor.

So no. It won't be forcing to always be there. You're just limiting, placing a ceiling if you will. And that's where the benefit comes from. 
Because (if you run it stock) the voltages are "tabled" according to clocks (the higher the clocks, the higher the voltage, lower clocks translate to lower voltages and etc). 

Taking an i9 13900KS as example...
It boosts to 6.0Ghz in single-core (the "Max Turbo Frequency"). To get there, it requires more voltage and, even if it's just for one core, all of that power is offered, at once.
If instead you sync all the P-cores, and set to 5.6 Ghz (which is its stock "out of the box" all P-cores clock), it won't boost there anymore and will not reach such silly high voltage (plus, you also get lower temperatures). 

It's how it should have been (IMO), like correcting it (and restricting that). For which you actually don't get lower performance in gaming or most things really. 

Not saying that this will save your CPU forever, or that it won't degrade ever again (given the latest news, it seems deeper than that). But, no doubt, you're already cutting the worst and biggest offender, and easing things a whole lot.

Check Buildzoid's video, and how that single-core boost goes over 1.5v... (insane how it happens, and that's supposedly "normal" for Intel 13th/14th gen!! )....
Picking a part at 12:13 time of video, just keep watching:

 

Edited August 2, 2024 by LucShep

[Panzerlang]

Well, I don't have a clue how to "synch all cores" in my Gigabyte Z790. I've googled it, nada.

Edit: I believe it's the "Enhanced Multi-Core Performance" in my BIOS, already set to "Disabled" per the list above. I guess disabling it locks/synchs the cores.

Edited July 31, 2024 by Panzerlang

[LucShep]

8 minutes ago, Panzerlang said:

Well, I don't have a clue how to "synch all cores" in my Gigabyte Z790. I've googled it, nada.

which model?

[Panzerlang]

Just now, LucShep said:

which model?

Aorus Master.

[Panzerlang]

Running the all-core CPU-Z stress-test it shows the cores running at 5200-MHz to 5300. They hit 87c.

[Panzerlang]

And now, running Prime-95 again, they max out at 4900MHz. What fckery is this? Lol.

Well, none of this makes any sense. CPU-Z test pushes the cores to 5700 and 95c, Prime-95 tops out at 4900 and 82c. 
Max power in both cases is 253w however.

When this Intel finally blows up I'll be going AMD for the first time in my 30-year journey with PCs. Lol.

[kksnowbear]

16 minutes ago, Panzerlang said:

Well, none of this makes any sense.

Most likely the two utilities are just using different types of loads (for example AVX vs not).

I believe you can configure, at least to some extent, the testing that P95 does. Don't know about CPUZ stress test as I don't use CPUZ for that. 

Edited July 31, 2024 by kksnowbear

[LucShep]

49 minutes ago, Panzerlang said:

Aorus Master.

Z790 Aorus Master manual
https://download.gigabyte.com/FileList/Manual/mb_manual_intel700series-bios_e.pdf?v=aceb9fb3f69cc73ea6b2fddd6a6f34ed

I'm mostly used to ASUS and MSI (what I'm used to for OC settings)... I see it's very different with Gigabyte in this aspect here.
I'm also not seeing the direct option to sync cores(?) and neither where to place a single value for all at once.  LOL
(maybe someone else can chime in?)

I do see this in your board's manual:


So interpreting as I read it.........

If you enter BIOS and go to:

"Advanced mode" (TWEAKER)
...
"Advanced CPU Settings"

(scroll down)

Turbo Per Core Limit Control ---- AUTO >>>> MANUAL
...
Turbo P-Core 0 Ratio Limit  ----   Auto >>>>> value of clock per core // for example "56" for 5.6Ghz, as for i9 13900KS all P-core max clock, as said previously
Turbo P-Core 1 Ratio Limit  ----    ´´     ´´     ´´
Turbo P-Core 2 Ratio Limit  ----    ´´     ´´     ´´
Turbo P-Core 3 Ratio Limit  ----    ´´     ´´     ´´
Turbo P-Core 4 Ratio Limit  ----    ´´     ´´     ´´
Turbo P-Core 5 Ratio Limit  ----    ´´     ´´     ´´
Turbo P-Core 6 Ratio Limit  ----    ´´     ´´     ´´
Turbo P-Core 7 Ratio Limit  ----    ´´     ´´     ´´


Again, this is how I interpret it as I'm diagonally reading the BIOS manual. But I could be wrong(!), it could be other settings...
I wish to not make you do something wrong!

Again, anyone knowing how the particular motherboard settings are, please chime in.

PS: my dislike for Gigabyte boards increased even more! LOL   (Go ASUS and MSI !!)

Edited August 1, 2024 by LucShep

[kksnowbear]

I believe it's possible that the option to set All Cores isn't "exposed" unless/until you select one or another form of overclocking (i.e. XMP or Manual, but sometimes "Expert" or similar). Then you see All Core or Per Core.

That's generally speaking, and I would also caution I am not intimately familiar with the Gigabyte BIOS at all. I'm not where I can really look into it ATM.

[Panzerlang]

49 minutes ago, kksnowbear said:

Most likely the two utilities are just using different types of loads (for example AVX vs not).

I believe you can configure, at least to some extent, the testing that P95 does. Don't know about CPUZ stress test as I don't use CPUZ for that. 

 

Yeah, I figure P95 isn't as brutal as CPU-Z.
"Performance CPU Clock Ratio" at 57 obviously limits (limits?! Lol) the speed to 5700 on the base clock of 100. No hiccups in either test with power limited to 253w but the package temp still hits 95c. I think I'll reduce it to 55, though I doubt DCS would push useage that hard anyway.

35 minutes ago, kksnowbear said:

I believe it's possible that the option to set All Cores isn't "exposed" unless/until you select one or another form of overclocking (i.e. XMP or Manual, but sometimes "Expert" or similar). Then you see All Core or Per Core.

That's generally speaking, and I would also caution I am not intimately familiar with the Gigabyte BIOS at all. I'm not where I can really look into it ATM.

What I was able to find was a post asking what "Synch all Cores" on ASUS BIOS translated into on Gigabyte BIOS and the answer was "Enhanced Multi-Core Performance".

The overall thrust of the stuff in the original post was to fix motherboards (all brands apparently) from allowing unlimited wattage AND voltage-spiking single cores in turbo mode.

[CptBligh]

15 hours ago, Panzerlang said:

And now, running Prime-95 again, they max out at 4900MHz. What fckery is this? Lol.

Well, none of this makes any sense. CPU-Z test pushes the cores to 5700 and 95c, Prime-95 tops out at 4900 and 82c. 
Max power in both cases is 253w however.

When this Intel finally blows up I'll be going AMD for the first time in my 30-year journey with PCs. Lol.

Prime95 uses AVX instructions.  Your BIOS should have a setting for AVX offset with reduced clocks to keep it from cooking itself with those instruction sets so that's why you have lower max clocks in P95.

[kksnowbear]

10 minutes ago, CptBligh said:

Prime95 uses AVX instructions.  Your BIOS should have a setting for AVX offset with reduced clocks to keep it from cooking itself with those instruction sets so that's why you have lower max clocks in P95.

P95 is configurable to not use AVX (in every version I've used, anyway).

Unless I'm mistaken, the default in BIOS is usually zero for the AVX offset, so unless you change it, it doesn't lower AVX clocks...I could be wrong here though.

Edited August 1, 2024 by kksnowbear

[CptBligh]

1 minute ago, kksnowbear said:

P95 is configurable to not use AVX (in every version I've used, anyway).

I'm responding to the CPU behavior he questioned that is related to AVX being enabled by default in P95 and lower AVX offset clockspeeds from his BIOS. 

"Selecting the Instruction Set Extension

By default, Prime95 automatically selects the newest instruction set extension, such as AVX, AVX2, or even AVX-512. In order to change this behavior, Prime95 needs to be started and completely shut down again once. This will create the local.txt file. In it, exclusions are assigned a value of 0, whereas the code path that’s to be tested is assigned a value of 1. If you aren’t clear as to which SSE version is supported by your CPU, both can be set to 1. Prime95 will choose the correct fallback."

[kksnowbear]

5 minutes ago, CptBligh said:

I'm responding to the CPU behavior he questioned that is related to AVX being enabled by default in P95 and lower AVX offset clockspeeds from his BIOS.

As I already did, several posts earlier:

15 hours ago, kksnowbear said:

Most likely the two utilities are just using different types of loads (for example AVX vs not).

I believe you can configure, at least to some extent, the testing that P95 does. Don't know about CPUZ stress test as I don't use CPUZ for that. 

Edited August 1, 2024 by kksnowbear

[CptBligh]

5 minutes ago, kksnowbear said:

As I already did, several posts earlier:

 

And I added further clarification to your caveats of "most likely' and "I believe" and how it ties into other BIOS settings.  I'm backing up what you posted, so not sure why you're trying to be argumentative.

[kksnowbear]

Prime95 doesn't have to use AVX instructions, and you don't have to change any files so it won't.  It's in the dialogue box for testing, and can be changed every time, without ever editing any file.  (What you're talking about is changing the default when it first opens; again, this is not required to change the behavior for testing).

The AVX BIOS offset would (likely) be set for zero, if I'm not mistaken, so if that were causing the different clocks, he would know that he changed it.  And if he knows that, well...he probably knows how the rest works (granted I'm assuming).  The BIOS manual doesn't appear to specify, so without the board in front of me, I couldn't say....hence:

I say most likely and I believe not because I think I'm wrong but because I haven't specifically done things like looking at his motherboard BIOS (as one example).  I don't necessarily have time to rtfm for everyone who hasn't

Don't get so excited.

Edited August 1, 2024 by kksnowbear

[Panzerlang]

I have it set at max 5.5GHz, temps have gone no higher than 85c with tests pushing the clock to 5.3GHz (neither test will push it to run faster) and power at 217w (max is 253w). I'm happy with that.

Thanks for the input guys.

Going to give Cinebench a go next.

[Panzerlang]

Still weird behaviour. At idle both CPU-Z and HWMonitor show the CPU min and max speed at 5.5GHz and wattage around 20w. When I run Cinebench (supposedly a hard-core tester) the values fluctuate between min 4.8GHz and max 5.2GHz, with power at 207w. Maybe I have a false memory but back when I used to OC my rigs a test (Prime-95 mostly) would always push the CPU to the max speed I'd set it at in the BIOS.

Cinebench is a 10-minute test. On the CPU multicore test it scored 1831. Max temp reached was 83c.

I'm completely unfamiliar with a CPU that idles at 5.5GHz with a power-draw of around 20w but will then run at a max of 5.2GHz and a min of 4.8GHz under test.

On top of that, even though the test is finished but with the app window still open, the CPU is still running at 207w and 83c. Now I've closed the app window, no change. I'm going to reboot the PC...

[Panzerlang]

...aaand...at idle, power is around 25w (browser open), temp fluctuates around 40c, CPU utilization around 3%.

Something is bent. The CPU should return to idle after a task is finished but it doesn't, it maintains whatever state it was last in. Maybe it's broken or the BIOS is fubar.

[Panzerlang]

DCS running on the monitor (Mig21 over Caucasus), full-screen, at 3480x2160, the CPU runs at full speed (5.5GHZ) but only 23% utilization. GPU (4090) is at 40% utilization. It has gone back to idle after shutting down the game though.